Turn On The Lights

Deploying advanced IP cameras in the challenging light conditions of rolling stock environments

• By Kevin Shen
• Nov 01, 2016

Whenever a surveillance system is trying to smooth video, light plays a very important role. Most security professionals have acquired important video only to see that the resulting footage is over or under exposed because the camera is unable to adapt to the challenging light environment. Trains are noted for being a particularly difficult surveillance environment and several unique challenges arise and must be overcome in order to achieve consistently smooth video footage. For example, a fast-moving train will often experience severe light fluctuations when traveling through tunnels, open air and shade, or when another train passes by. Yet these challenges are not limited to what is happening outside of the train, but also include onboard challenges such as the lights being switched on and off, and doors opening and closing. In order to ensure clear imagery on trains, a tailor made solution is required.

TECHNOLOGY OVERVIEW AND THE CHALLENGES

Auto Exposure (AE) is a technology that has been around for many years. It adjusts the shutter, IRIS, and gain based on the ambient light captured by the sensor. The AE algorithm uses an image taken in ideal light conditions and stores it as a reference image. If the current image brightness is higher or lower than the reference image, the AE algorithm will adjust the current image to make it the same as the reference image. However, if the change of brightness affects a small part of the overall image, or if there is a significant light change in a small area, the traditional AE algorithm will readjust the whole image, which is a problem that must be addressed before the AE feature can be effectively deployed on trains.

Two of the most difficult challenges when using the AE function on trains are caused by a combination of environmental and human factors. In a stable light environment, the AE performs its function well but in the complex light environment of a train, the standard AE algorithm cannot achieve the standards required. Due to space restrictions onboard trains, most cameras that are deployed will be compact, and are unable to accommodate an adjustable IRIS.

For this reason, the AE feature that we will consider in this article is related to the shutter and gain.

A typical problem that is encountered on trains is when the AE algorithm compensates for adjustments in light when it is not necessary. An example is when passengers who are wearing light-colored clothing enter the carriage and increase the brightness of the image causing the AE algorithm to reduce the image brightness. If those passengers then leave the carriage, the AE algorithm will adjust the image again because the image has become darker. A similar situation occurs when the light conditions inside the carriage change due to the train entering or exiting a tunnel or passing by some buildings that block out the light. The challenge of the current AE algorithm is to keep it functioning as it was intended, but also to avoid any unnecessary adjustments.

SOLUTION FOR THE PASSENGER

First, light conditions are very stable in cars with a sustained light source. These cars will not experience significant light fluctuations internally as the train moves. As a result, the AE algorithm does not need to be that sensitive to the light fluctuations that occur outside the train.

Another factor that must be considered is whether a light change within a small part of the image should be ignored. In general, the AE function will compensate when there is a change in the lighting environment. Most AE algorithms are designed to modify the image based on the overall image change. However, as the light source is generally quite stable on a train, a slight change in brightness would be attributed to a change in the objects present in the scene and would not trigger the AE adjustment for this scenario. The AE feature would still compensate if the image brightness changed considerably to ensure clear imagery.

We tested two AE algorithms by installing two cameras in a car consist and capturing video as the train entered a tunnel. The first incorporates a traditional AE algorithm while the second has a new AE algorithm developed by Moxa.

TAMPERING DETECTION FUNCTION

Fluctuating light conditions not only affect image quality, but also impact some functions that help enhance security. The tampering function has traditionally been unstable when deployed in onboard environments. However, this is a problem that must be overcome for users who require excellent video quality and a reliable security system installed in their onboard environment.

The camera tampering feature automatically detects when a camera is being tampered with and issues an alert to the user. This feature performs a comparative analysis based on a digital reference image taken by the camera during a period of ideal brightness. The amount of time from detecting a possible event to issuing an alert can be split into three phases. First, there is a learning stage when a possible tampering event is detected. Second, is the detection stage, when the algorithm continues to detect if there is any significant image change based on the reference image and also uses the new image to upgrade the detection sample. Finally, once a large change has occurred, the function will trigger the event alarm.

In order for the tampering algorithm to accurately determine whether a camera is being tampered with, a ratio needs to be established between the expected amount of change to the scene the camera would typically experience onboard a train and when the camera is actually being tampered with.

As onboard trains are constantly changing environments a clear challenge presents itself if the user wants to deploy accurate tampering algorithms onboard a train. Changes to image brightness can be caused by many different factors, and users do not want to receive a tampering alert for normal occurrences such as the train entering a tunnel or passengers standing in front of the camera. In some scenarios, it is difficult for the algorithm to differentiate between when actual tampering of the camera is taking place and a normal image change onboard a train occurs.

CHALLENGES FOR DEPLOYMENT

It is important that the tampering algorithm judges correctly if it should or should not send an alert. False alerts are when a normal event triggers the alarm, and missed alerts are when a real tampering event occurs but the camera fails to send an alert. Several challenges need to be overcome so that users can avoid these two troublesome scenarios.

• When a camera is being tampered with the color of the scene will change; therefore, a change of color should be counted as one of the factors for triggering the tampering alarm. Due to passenger movement or when the train moves through different environments the scene on a train will experience a change in colors. In either case, the camera’s algorithm will send a false alert because it believes that a tampering event is taking place.
• Another factor to decide whether to trigger the tampering alarm or not is when a small area of a scene undergoes significant light change. However, in a very simple or a very complex environment, it might be difficult to judge if this is an actual tampering event.
• Different scenarios require different parameters for triggering the tampering alarm. For example, frequent and large changes to the image should be expected in a crowded car. However, there will be places on the train where the image is more stable. Thus, the feature needs to be flexible in order to cope with different situations. After considering the three challenges above, an effective algorithm will be able to judge whether a camera is being tampered with as well as a human operator could judge. In order to increase the accuracy of the tampering function, below are some solutions to the aforementioned challenges.

HOW THE CHALLENGES CAN BE OVERCOME

A variety of factors must be considered to determine whether the camera is being tampered with or not. The camera will not only consider the change in brightness of the image, but also the contrast and other relevant factors.

Several contributing factors allow the camera to get as close to a human’s judgement as possible. The sensitivity level of different parts of the scene can be fine-tuned to better suit different environments, as shown in Fig. 5 below. The algorithm can consider overall image changes and also partial image changes at the same time. The tampering alarm should only be triggered if the number of partial changes is sufficient to influence the overall change.

Maintaining excellent image quality onboard trains is not a simple task. Several measures and countermeasures need to be considered to meet fluctuating light conditions, and to ensure that the tampering alarm is not triggered accidentally and still functions properly when a camera tampering event occurs.

This article originally appeared in the November 2016 issue of Security Today.